Hi Ralf, Ralf Wildenhues wrote: > What is unclear about the following, quoting from > http://www.gnu.org/software/autoconf/manual/html_node/Present-But-Cannot-Be-Compiled.html > ? > > | The proper way the handle this case is using the fourth argument > | (see Generic Headers): > | > | $ cat configure.ac > | AC_INIT([Example], [1.0], [bug-example@xxxxxxxxxxx]) > | AC_CHECK_HEADERS([number.h pi.h], [], [], > | [[#if HAVE_NUMBER_H > | # include <number.h> > | #endif > | ]]) > >> I am going to take a wild guess that >> I replace this line from snprintfv.m4: >> >> AC_CHECK_HEADER(wchar.h) >> >> with this: >> >> AC_CHECK_HEADER([wchar.h], [], [], >> [[#if HAVE_WCHAR_H >> # include <wchar.h> >> #endif >> ]]) > > That wouldn't make any sense. A prerequisite header would typically be > a different one. That is (almost) exactly the example from the Present-But-Cannot-Be-Compiled chapter, modulo s/number/wchar/. I agree that it makes no sense. That's why I forwarded the issue to the autoconf list. If you were meaning to show an example of "pi.h" needing, but not self-including, "number.h" and how to get around it, you need some more exposition on that. It _is_ unclear what you are getting at. Perhaps just clarify by breaking it into two "AC_CHECK_HEADER()" invocations. Not only do I not know the nitty-gritty implementation details of AC_CHECK_HEADERS(), I also have a strong desire for not learning them. :) Understanding how your fourth argument works there means knowing that every listed header is completely handled before starting on the next listed header. If you were to emit all the #define HAVE_WHATEVER_H at the end, then your example here breaks. (I didn't want to figure that out.) >> Since reasonably modern system headers should be self-sufficient >> (you should 'a been there in the seventies and eighties!!), > Welcome back to the seventies and eighties then, I guess. No. Back then, very few headers were self sufficient and programmers were expected to learn the proper order and dependencies of system header files. Part of the programmer hazing rituals, I guess. So, with Tomaz' input, I now know the fix. Snprintfv needs some more configury magic for wint_t. It presumes that HAVE_WCHAR_H implies that any system defined wint_t is visible if you #include <wchar.h>. So, OS/X is "special" here. I suggest adding a specialized AC_CHECK_WINT_T macro that deals with this Darwinism and I'll hack up my configury magic to ensure that <ctype.h> gets rolled into the required headers before the existence of wint_t is tried. > I *then* looked at the DarwinPorts portfile, which runs the standard > "./configure ; make" but before doing so, defines CPPFLAGS to be > "-D_WINT_T". This gets around the "lipopts.c" compile It would be "really nice" if folks who hack around such issues would send me a note. I could have fixed this as long ago as someone first discovered this issue. :( Are there any other mods in the "portfile", Tomaz? Thanks Ralf & Tomaz for your feedback. Regards, Bruce _______________________________________________ Autoconf mailing list Autoconf@xxxxxxx http://lists.gnu.org/mailman/listinfo/autoconf
